2.1. Introduction

The enzyme super family of P450 cytochromes provides some of the most sophisticated catalysts of drug biotransformation reactions. These enzymes catalyze a wide variety of oxidative and reductive reactions and demonstrate activity on a chemically diverse variety of substrates. Moreover, a small subset of the CYP enzymes is responsible for the majority of drug-metabolizing reactions.

Several aspects of these enzymes, such as the rate and site of metabolism, inhibition, induction, and the selectivity of the various isoforms, are relevant in the lead optimization process during the discovery of new therapeutic agents.

Although there is a large amount of information on the functional role of these enzymes, knowledge of their dynamic motion is still lacking. In vitro screens for inhibition and metabolic stability can provide some of the information required in drug discovery. The experimental elucidation of the site of metabolism is usually a highly resource-demanding task, requiring several experimental techniques and consuming a considerable amount of compound. Nevertheless, the recognition of the site of metabolism is a great help in designing new compounds with improved pharmacokinetic profiles and in avoiding the presence of toxic metabolites by chemically protecting the metabolic labile moieties in the drug candidate. Another use of metabolism site prediction is in designing pro-drugs where the compound needs to be metabolized in order to become active.

Using the 3D structure of a compound, MetaSite [1-11] is a computational procedure specifically designed to predict the site of metabolism of xenobiotics and the derived metabolites.

The MetaSite procedure is completely automated, easy to apply, and does not require user assistance. All of the work can be submitted and handled in batch queue. MetaSite uses GRID-based representations of the CYP enzymes that have been pre-computed and are stored within the software. The calculations of charges and reactivities, pharmacophoric recognition, descriptor handling, and similarity computation are carried out automatically once the 3D structure of the compounds has been provided. Figure 2-1 shows a schematic representation of the MetaSite workflow.

2.2. Protein Treatment

The GRID [12] force field is used to produce Molecular Interaction Fields inside the cytochrome active-site. Program GRID is calibrated in a water environment to obtain chemically specific information about a (macro)molecule (in this case the human cytochrome). An electrostatic potential does not normally allow favorable binding sites to be differentiated for a primary, secondary, or a tertiary amine cation, for pyridinium, or for a sodium cation, and the GRID method is an attempt to compute analogous potentials which do have some chemical specificity. The object used to measure the potential at each point is given the generic name Probe. Many different Probes can be used on the same macromolecule one after another, and each represents a specific chemical group. A great deal of chemically specific information can therefore be accumulated concerning the way in which the macromolecule might interact favorably with other ligand molecules.

The molecular interaction fields (MIF) in the binding site of the cytochromes were obtained using the flexible mode in GRID. With this flexible option, some of the amino acid side-chains can automatically move in response to attractive or repulsive interactions with the chemical probe. The side-chain flexibility in GRID can mimic the amino acid movements that occur in the CYPs active site to accommodate different substrates according to their size, shape, and interaction pattern.

The MIFs obtained from cytochrome enzymes are subsequently transformed and simplified. In a cytochrome, where the catalytic reaction takes place, all of the 3D map information can be compressed and refers to the catalytic centre of the enzyme, that is, the oxene atom of the protoporphyrin group.

The selected 3D interaction points are used to calculate enzyme fingerprints using the GRIND technology. For each CYP-Probe interaction map, this approach transforms the interaction energies at a certain spatial position (the MIF descriptors) into a number of histograms that capture the 3D pharmacophoric interactions of the flexible protein. Such histograms are called correlograms (Figure 2-2). The correlograms represent the distances between the reactive center of the cytochrome (the oxene in the heme moiety) and the different chemical regions inside the enzyme active site.

2.3. 3D structure of substrates, and fingerprint generation

The majority of CYP substrates contain flexible moieties. Since conformation is relevant to recognition of the ligand substrate and binding to the CYP, and has a direct impact on the outcome of the method, each substrate is modelled by using a population of diverse low-energy minimum conformations, provided automatically by the computational procedure. Each of the 3D conformers is used for the fingerprint calculation.

The descriptors developed to characterize the substrate chemotypes are obtained from GRID probe pharmacophore recognition. The charges computed by GRID are used to derive a 3D pharmacophore based on the molecular electrostatic potential (MEP) around the substrate molecules.

Moreover, all of the substrate atoms are classified into GRID probe categories depending on their hydrophobic, hydrogen-bond donor, or acceptor capabilities. Their spacial distances are then binned and transformed into clustered distances. One set of descriptors is computed for each atom type category: hydrophobic, hydrogen-bond donor, hydrogen-bond acceptor, and charged, thus yielding a fingerprint for each atom category in the molecule. The distances between the different atomic positions classified using the previous criteria are then transformed into binned distances. In this case, the distances between the different atoms are calculated and a value of one or zero is assigned to each bin distance, respectively indicating the presence or absence of such a distance in the substrate (Figure 2-3).

2.4. Substrate-CYP enzyme comparison: the Recognition component

Once the protein interaction pattern is translated from Cartesian coordinates into distances from the reactive centre of the enzyme, and the structure of the ligand has been described with similar fingerprints, both sets of descriptors can be compared. The complementarity of the hydrophobic MIF distances,charge, H-bond donor and acceptor distances for the protein and the substrate, are all computed using Carbó similarity [13]indices (Figure 2-4). The prediction of the site of metabolism is based on the hypothesis that the distance between the reactive centre on the protein (oxene atom in protoporphyrin group) and the interaction points in the protein cavity (GRID-MIF) should correlate with the distance between the reactive centre of the molecule (i.e. positions of hydrogen atoms and heteroatoms) and the position of the corresponding interacting atom types in the molecule.

A similarity score is assigned to each atom in each substrate. Due to the mechanism of computation [8], the score is proportional to the exposure of such substrate atoms toward the reactive heme, and represents the accessibility component.

The accessibility component, called Ei, represents the recognition between the specific CYP-protein and the ligand substrate when the ligand is positioned in the CYP-protein and exposes the atom i toward the heme. This depends on the ligand 3D structure, conformation, stereochemistry, and on the 3D structure and side-chain flexibility of the CYP-enzyme. Thus the Ei score is proportional to the exposure of the ligand atom i toward the heme group of a specific CYP-enzyme (Figure 2-4).

2.5. The Reactivity component

Cytochromes P450 catalyze oxidative and reductive reactions.Oxidative bio-trasformations are more frequent and include aromatic and side-chain hydroxylation, N-, O-, S-dealkylation, N-oxidation, sulfoxidation, N-hydroxylation, deamination, dehalogenation, and desulfuration. The majority of these reactions require the formation of radical species, which is usually the rate determining step for the reaction.

When Ri is the reactivity of atom i in the appropriate reaction mechanism, it represents the activation energy required to produce the reactive intermediate. It depends on the ligand 3D structure and on the mechanism of reaction. Therefore, Ri is a score proportional to the reactivity of the ligand atom i in a specific reaction mechanism.

Furthermore, for each reaction mechanism, Ri does not depend on the P450 enzyme, but is only related to the molecular topology and 3D structure.

2.6. Computation of the probability of site of metabolism

Three driving forces control the site of metabolism for substrates of human CYP enzymes. Calculations show that for all of the atoms of the test molecules, the probability of being the site of metabolism depends firstly on the enzyme accessibility, called Ei, secondly on the chemical reactivity, called Ri, and lastly on the reaction mechanism, called Mi.

Once these three components are calculated, the site of metabolism can be described by a probability function P_SM (Probability for the Site of Metabolism) reported in Equation (2), which is correlated to and can be considered to be an approximation of the free energy of the overall process.

          P_SMi = Ei * Ri * Mi           Equation (2)

where:

• P_SMi is the probability of an atom i being the site of metabolism catalyzed by the CYP-heme group;

• Ei is the accessibility of atom i to the Heme;

• Ri is the reactivity of atom i in the actual mechanism of reaction.

• Mi is the relative probability of a reaction mechanism under consideration occurring.

Ei is the recognition score between the CYP-protein and the ligand when the ligand is positioned in the CYP-protein and exposes the atom i towards the heme. It depends on the 3D structure, conformation, and chirality of the ligand, and on 3D CYP-protein structure. The Ei score is proportional to the exposure of atom i to the heme group. Similarly, Ri is the reactivity of atom i in the appropriate reaction mechanism, and represents the activation energy involved in producing the reactive intermediate. It depends on the 3D structure and topology of the ligand.

When different reaction mechanisms are possible, Mi is the relative probability of each mechanism occurring. Mi can be also considered to be a selectivity factor as it is able to discriminate between reaction mechanisms in different CYP isoforms.

For the same ligand and the same cytochrome, the P_SM function assumes different values for different ligand atoms depending on the Ei, Ri and Mi components. When a ligand atom i is well exposed to the reaction center of the heme (Ei has a high score), but its reactivity is very low (Ri reports a very low score), the probability of metabolism in atom i will be very low or zero. Similarly, when a ligand atom i is very reactive in the mechanism considered (Ri reports a high score), but atom i is not exposed to the reaction center of the heme (Ei has a very low score), the probability of metabolism in atom i will be close to zero. Therefore, to be the site of metabolism, an atom i should possess significant accessibility and reactivity components related to the heme. However, when two atomic positions i and j show similar values of each of the single components Ei or Ri (or similar value of the product of Ei * Ri), then the reaction will favor atom i when the mechanism component Mi is greater for i than for j.

2.7. The reaction mechanism component

Two common reactions when molecules are exposed to CYPs are aniline oxidation and benzylic oxidation. Some CYPs show no preference for these two reactions i.e. when Ei and Ri components are similar, the Mi component for these two reactions is also very similar. However, there are CYPs showing a strong preference for certain reactions. Accordingly, when Ei and Ri components are similar, the Mi component can be very different. Therefore, for 1A2, which shows a similar Mi component for aniline and benzylic oxidations, the Ei and Ri components are responsible for the ranking of aniline or benzylic oxidation. Conversely, for 3A4, which shows different Mi components for aniline and benzylic oxidations, the Mi component is more important and the ranking of SoM can be strongly influenced by the Mi factor.

2.8. Consensus SoM prediction models

The consensus SoM prediction models perform an averaging process on the results coming from the independent predictions performed with the enzymes that are most relevant for the specific organ in the human body.

The current version of the software implements the following consensus models:

• LIVER (CYP2C9, CYP2D6, CYP3A4)

• SKIN (CYP1A1, CYP2E1, CYP3A5)

• BRAIN (CYP1A1, CYP2D6, CYP3A4)

The consensus SoM prediction relies on how frequently the same site of a fixed substrate structure is selected among the individually predicted most probable sites; it doesn't therefore directly depend on the Psm numerical values. The consensus is obtained by assigning the same relative weights to each CYPs, although we know that some CYP isoforms are more abundant than others depending on the particular organ. The relative weights will be modified only when we have a clear indication of the percentage of CYP involved in the metabolism of each molecule.

2.9. Relative retention times

The estimated relative retention time (RRT) of the metabolites is based on the hydrophilic/lipophilic balance of each metabolite when compared to the parent substrate.

In the cases where experimental retention times were reported together with the pH of the chromatographic column, a good correlation of the estimated RRT with experimental data has been observed.

The predicted RRT therefore provides valuable information when discriminating between metabolites with equal mass during HPLC separation phase.

The RRT values represent the following quantity:

RRT = (RT(metabolite)-RT(substrate))/RT(substrate).

Metabolites with a retention time higher than that of the substrate are therefore expected to have a positive RRT (and vice versa).

Please also note that the estimated RRT depends on the pH of the chromatographic column. By default, the RRT chart in the MetaSite GUI refers to an acid environment, but RRT values referring a neutral or basic environment are produced as well.

2.10. Metabolite Generation

Together with "hot spot" SoM predictions, MetaSite can also predict the structure of the metabolite(s) formed after the metabolic biotransformation(s).

The procedure produces a full set of metabolite structures based on a number of possible metabolic reactions. The metabolites formed are ranked according to the MetaSite Site of Metabolism predictions, yielding a ranked list of metabolites in conjunction with their monoisotopic mass and lipophilicity values. The lipophilicity values are then used to predict the relative retention time of the metabolites.

2.11. Mechanism-Based inhibition

Mechanism-based (MB) inhibitors are chemically reactive substrates or metabolites capable of irreversibly reacting with the cytochrome enzymes, resulting in their inactivation. The reactive substrates can sometimes be formed during the cytochrome catalytic mechanism. The experimental evidence of this inactivation is time-dependent inhibition. The more reactive the metabolite formed, the less protein is available for subsequent catalysis.

Any compound can be a potential MB inhibitor if it posseses three main characteristics:

• it is a good CYP substrate;

• it contains appropriate reactive molecular moieties;

• it exposes these reactive moieties to the heme.

2.12. References

1. Zamora I. Site of Metabolism Predictions: Facts and Experiences. In Antitargets Prediction and Prevention of Drug Site Effects. Vol 38: Methods and Principles in Medicinal Chemistry, Volume Editors: R.J. Vaz and T. Klabunde, Series Ed.: R. Mannhold, H. Kubinyi, G. Folkers, Wiley-VCH Verlag GmbH and Co. KHaA, 2008, pp: 247-262

2. Aristei Y, Cruciani G, Clementi S, Carosati E, Vianello R, Benedetti P. MetaSite: Understading CYP antitarget modeling for early toxicity detection. In Antitargets Vol 38: Methods and Principles in Medicinal Chemistry, Volume Editors: R.J. Vaz and T. Klabunde, Series Ed.: R. Mannhold, H. Kubinyi, G. Folkers, Wiley-VCH Verlag GmbH and Co. KHaA, 2008, pp: 277-2901.

3. Ahlström MM, Ridderström M, Zamora I., CYP2C9 structure-metabolism relationships: substrates, inhibitors, and metabolites; J. Med. Chem. 2007 Nov 1;50(22):5382-91.

4. Ahlström MM, Ridderström M, Zamora I, Luthman K., CYP2C9 structure-metabolism relationships: optimizing the metabolic stability of COX-2 inhibitors; J. Med. Chem. 2007 Sep 6;50(18):4444-52.

5. Kjellander B, Masimirembwa CM, Zamora I. Exploration of enzyme-ligand interactions in CYP2D6 & 3A4 homology models and crystal structures using a novel computational approach; J. Chem. Inf. Model. 2007 May-Jun;47(3):1234-47.

6. Zhou D, Afzelius L, Grimm SW, Andersson TB, Zauhar RJ, Zamora I.Comparison of methods for the prediction of the metabolic sites for CYP3A4-mediated metabolic reactions; Drug Metab. Dispos. 2006 Jun;34(6):976-83.

7. Cruciani G, Aristei Y, Vianello R, Baroni M. GRID-Derived Molecular Interaction Fields for Predicting the Site of Metabolism in Human Cytochromes In Molecular Interaction Fields. Applications in Drug Discovery and ADME prediction Vol 27 Methods and Principles in medicinal Chemistry, Ed. G. Cruciani, Series Editors: R. Mannhold, H. Kubinyi, G. Folkers, Wiley-VCH Verlag GmbH and Co. KHaA, 2006, pp: 277-290.

8. Cruciani G, Carosati E, De Boeck B, Ethirajulu K, Mackie C, Howe T, Vianello R. MetaSite: Understanding Metabolism in Human Cytochromes from the Perspective of the Chemist; J. Med. Chem.2005; 48:6970-6979.

9. Berellini G, Cruciani G, Mannhold R. Pharmacophore, drug metabolism, and pharmacokinetics models on non-peptide AT1, AT2, and AT1/AT2 angiotensin II receptor antagonists; J. Med. Chem.2005 Jun 30; 48(13):4389-99.

10. Cruciani G, Aristei Y, Vianello R, Baroni M. GRID-Derived Molecular Interaction Fields for Predicting the Site of Metabolism in Human Cytochromes.In Molecular Interaction Fields: Applications in Drug Discovery and ADME Prediction; Cruciani, G. (Ed.);Copyright © 2005. WILEY-VCH Verlang GmbH & Co. KGaA, Weinheim.

11. Zamora I, Afzelius L, Cruciani G. Predicting Drug Metabolism: A Site of Metabolism Tool Applied to the Cytochrome P450 CYP2C9; J. Med. Chem. 2003; 46(12):2313-2324.

12. Goodford PJ. A Computational Procedure for Determining Energetically Favorable Binding Sites on Biologically Important Macromolecules; J. Med. Chem. 1985; 28:849-857.

13. Amat L, Carbó-Dorca R. Fitted Electronic Density Functions from H to Rn for Use in Quantum Similarity Measures: Cis-diamminedichloroplatinum(II) Complex as an Application Example; J. Comp. Chem. 1999; 20(9): 911-920.